Image Segmentation Fusion with Normal Vectors for Edge Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Deep learning algorithms based on convolutional neural networks suffer from poor segmentation accuracy due to partially missed segments, while traditional edge detection and plane estimation methods require high-quality images and struggle with blurred or irregular edges.
Innovation Solution
An image segmentation method involving obtaining a preliminarily segmented image and a target normal vector image, followed by image fusion to enhance the segmentation process, using pre-trained models and normal vector extraction to improve accuracy and stability.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If deep learning algorithm based on convolutional neural network is used, then segmentation speed is improved, but segmentation accuracy deteriorates due to partially missed segmentation
Solution Approach 1:
The patent combines deep learning-based segmentation results with traditional edge detection and plane estimation methods. The fusion module integrates the preliminarily segmented image (from deep learning) with the target normal vector image (from traditional methods) to produce the final segmented image, thereby combining the speed advantage of deep learning with the accuracy advantage of traditional methods.
Solution Approach 2:
The patent introduces a fusion module as an intermediary that processes both the preliminarily segmented image and the target normal vector image. This intermediary component reconciles the differences between the two methods by weighting and combining their respective information to produce an accurate final segmentation result.
2Measurement precision
If traditional algorithm based on edge detection and plane estimation information is used, then segmentation accuracy is improved for smooth regions, but adaptability deteriorates making it difficult to segment images with blurred edges or irregular edges
Solution Approach 1:
The patent merges the strengths of both approaches by using the deep learning method to handle complex edge cases (blurred or irregular edges) while using the traditional method to provide structural information (normal vectors) for accurate segmentation in smooth regions. The fusion module combines these complementary strengths.
Solution Approach 2:
The patent changes the parameter space by introducing normal vector information as an additional dimension for segmentation. Instead of relying solely on pixel intensity or edge gradients, the system uses normal vector images to represent surface orientation, which provides robust information for both smooth and irregular edges.
3Speed
If deep learning algorithm is used, then processing speed is improved, but reliability deteriorates due to poor segmentation effect
Solution Approach 1:
The fusion module acts as a feedback mechanism that refines the preliminary segmentation results. By incorporating normal vector information from traditional methods, the system provides feedback to correct and improve the segmentation results, thereby enhancing reliability while maintaining the speed advantage of deep learning.
Data Source
AI summary
The present disclosure provides an image segmentation method and apparatus, an electronic device, and a storage medium. The image segmentation method including: obtaining an image to be segmented; determining a preliminarily segmented image and a target normal vector image that are corresponding to the image to be segmented; and performing image fusion on the preliminarily segmented image and the target normal vector image to obtain a target segmented image.